Qwen3.6-Max-Preview: Smarter, More Refined, and Still Evolving

Qwen3.6-Max-Preview is a preview of the next proprietary model following Qwen3.6-Plus, strengthening world knowledge and instruction following while also improving agentic coding performance across multiple benchmarks. As it is still in the preview stage of development, further improvements are expected in later versions.

The extent of the performance gains was also presented in detail. Compared to the previous model, it recorded SkillsBench +9.9, SciCode +6.3, NL2Repo +5.0, Terminal-Bench 2.0 +3.8, SuperGPQA +2.3, QwenChineseBench +5.3, and ToolcallFormatIFBench +2.8.

Access paths were also disclosed together.

• Provided as a proprietary model hosted on Alibaba Cloud Model Studio
• Can be chatted with directly in Qwen Studio
• The API name is qwen3.6-max-preview
• Supports the preserve_thinking feature, allowing thinking content from previous turns to be retained during agentic tasks

In summary, this release is an update that previews Qwen's next-generation model, which strengthens coding agents, knowledge-based tasks, and instruction-following ability all at once. However, the key point is that this is not the final completed version, and Qwen has signaled further improvements in subsequent versions.